pingdotgg/t3code. All authors. Drafts included. Default branch main. 570 issue assessments and 925 PR assessments. Initial inventory: 570 open issues and 924 open PRs. Current assessed open inventory: 570 issues and 924 PRs. Final reconciliation: 2026-09-01T11:57:41.491617+00:00.
Request. Allow Git five minutes to remove dependency-heavy worktrees.
Audit finding. Main still gives git worktree remove only 15 seconds; the merged missing-worktree fix does not cover slow deletion. This two-file diff raises only that deadline and tests a removal beyond the old limit while leaving Git responsible for validation and cleanup. The reviewed head is clean and approved, Test, Check, and Release Smoke pass, and the native check is correctly skipped for this server-only change.
Recommendation. Keep open: work remains. Merge the reviewed head if its diff and checks remain unchanged.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Merged pr: PR #8076. Handles already-removed worktrees, not slow removal.
Comment: Discussion comment. Latest-head approval follows removal of the earlier unsafe filesystem-delete approach.
Check: GitHub check. Latest reviewed head passed required Test, Check, and Release Smoke; native analysis was skipped by its path gate.
Independent readiness check. The integrated change only raises the git-owned worktree removal deadline from 15 seconds to five minutes. It preserves path handling, the single --force argument, locked-worktree behavior and the already-removed-worktree branch. The delayed-removal test exercises the real driver past the old deadline.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head b457585d31192240a087610b8f5792d5ee800c7f. The synthetic merged tree was not built or executed.
Scope limit. Removal can still time out after five minutes. No unbounded cleanup or new destructive path is added.
Request. Avoid repeated full-prompt token scans when a large composer prompt has not changed.
Audit finding. The parser still rescans every nonempty prompt on main. Both touched files exactly match the patch base, the full patch applies cleanly to pinned main, all four required checks pass on its current head, and its review has no unresolved findings. The single-entry cache preserves fresh result objects and bypasses terminal-context calls.
Recommendation. Keep open: work remains. Merge the focused prompt-parser cache change.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Merged pr: PR #6602. The merged prompt-size validation does not cache composer parsing.
Check: GitHub check. Current-head Test passes, as do Check, Mobile Native Static Analysis, and Release Smoke in this run.
Independent readiness check. The integrated cache stores one exact prompt and only caches calls without terminal contexts. Cached outputs and returned results are separate arrays with fresh segment objects. The current token parser is a pure function of the same prompt, so no provider, workspace or mutable context is omitted from the cache key. Empty prompts reset the cache and changed prompts replace it.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head a983113dc32d0cd8ceffcf1855addb6309be2b94. The synthetic merged tree was not built or executed.
Scope limit. The cache retains one prompt until it is replaced or cleared. It does not cache an unbounded prompt history.
Request. Correct stale commands, paths, and documentation links.
Audit finding. The inspected main files still contain pnpm screenshot examples, the singular RPC_REQUIRED_SCOPE name, and references to a mobile-only Ghostty pin. Current scripts, source constants, and vendor paths confirm each replacement. This documentation-only diff is fully reviewed, has a head approval, and passes all four required checks. The separate Vercel failure asks for deployment authorization and does not cover changed application code.
Recommendation. Keep open: work remains. Merge the documentation-only change.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Check: GitHub check. Canonical latest-head checks: Check success, Test success, Mobile Native Static Analysis success, Release Smoke success.
Independent readiness check. The integrated patch changes documentation only. I checked every replacement against current main: the vp script names exist, RPC_REQUIRED_SCOPES is the source constant, the shared Ghostty pin and pairing script paths exist, and the new index targets resolve. All required canonical checks pass. The overall failure is only the non-required Vercel deployment-authorization check, and no marketing code changed.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head 7becfeae09392f09812e160418dfeeb969d25189. The synthetic merged tree was not built or executed.
Scope limit. This corrects listed stale references. It does not claim that every statement in the surrounding documentation has been revalidated.
Request. Show Send beside Stop during an active turn when the desktop composer has sendable content.
Audit finding. Main still hides Send on desktop during active turns while keeping it available on narrow web viewports. The patch removes only that viewport gate and retains the existing send, stop, and queued-input behavior. The full diff applies to pinned main, the primary-action implementation still matches the patch base, and current-head required checks and review pass.
Recommendation. Keep open: work remains. Merge the reviewed Send-button change.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Merged pr: PR #4781. This merged fix only makes Send reachable in mobile web layouts.
Check: GitHub check. The current head passes all four required checks, including the tests.
Comment: Discussion comment. Approval is for the unchanged current head.
Independent readiness check. The integrated change removes the viewport-only Send visibility gate. During ordinary active turns, Stop remains a type=button interrupt action and the existing type=submit Send button appears only with sendable content. Existing busy, disconnected and disabled guards remain. Pending questions and plan-follow-up actions still take precedence. No Queue/Steer or provider execution path changes.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head 3133379c482523775d4509b22129d27f06d847fd. The synthetic merged tree was not built or executed.
Scope limit. This changes desktop and responsive web action visibility. Native mobile behavior and provider steering support are unchanged.
Request. Find Homebrew-installed Node when launching a Linux or WSL backend without its interactive-shell PATH.
Audit finding. The shared SSH and WSL bootstrap still omits the fixed Linuxbrew bin directory. The two-line patch adds that directory to the existing engine-validated Node fallback without changing the installed-node fast path. Required CI and the current review pass, while the only red status is an authorization request for an unrelated Vercel marketing preview.
Recommendation. Keep open: work remains. Merge the focused Linuxbrew Node-discovery fix.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Check: GitHub check. Head 7f73186: Check, Test, and Release Smoke pass. Mobile Native Static Analysis is skipped. The unrelated Vercel marketing preview requires authorization.
Independent readiness check. The integrated patch adds the standard Linuxbrew bin directory to the existing guarded PATH fallback. A valid Node already on PATH still returns before any fallback changes, absent directories are ignored, duplicates are not added, and the existing engine check remains. The same generated helper serves SSH launch and the WSL backend.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head 7f73186893e9738d0e3b91e1660cc8e8c95a70a8. The synthetic merged tree was not built or executed.
Scope limit. The existing fallback ordering is unchanged. No new Linuxbrew-only WSL session was launched.
Request. Keep queries pending while a relay session is replaced or a saved environment starts connecting.
Audit finding. The merged query-retry change covers interruption, but main still exposes the session-loss race and the undesired initial supervisor state. This four-file follow-up closes both races with deterministic tests and retains terminal errors for real disconnects. Current-head Test, Check, and Release Smoke pass, native static analysis is correctly skipped for this scope, the merge is clean, and review findings are resolved.
Recommendation. Keep open: partial fix. Merge this focused connection-handoff follow-up.
Confidence high. Release: In stable source. PR readiness: Safe to merge now.
Merged pr: PR #8117. The earlier merged retry fix does not cover these two races.
Check: GitHub check. Current-head required Test passes. Check and Release Smoke also pass, and native static analysis is skipped for this non-native change.
Independent readiness check. The integrated query branch suppresses only EnvironmentRpcUnavailableError while its sampled connection state is still connected. That error originates when the live supervisor session is absent. The connection/session subscription then selects reconnecting or terminal behavior. Starting registry-owned supervisors as desired removes the transient manual-disconnect state without changing explicit disconnect behavior.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head b53887819be8d7cfead05e8ccbd1a89f1a62e059. The synthetic merged tree was not built or executed.
Scope limit. The check is scoped to query handoff. It does not hide command failures or change authentication errors.
Request. Allow a bounded Azure DevOps PR-list response larger than the shared one-megabyte output cap.
Audit finding. Main still applies the default output cap to the 100-row Azure PR-list path. This diff forwards a bounded 16 MiB override and tests a valid oversized response plus process-limit propagation. Latest Test, Check, and Release Smoke pass, native analysis is correctly skipped for these server-only files, and the focused source review found no remaining blocker.
Recommendation. Keep open: work remains. Merge this scoped Azure PR-list limit fix.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Comment: Discussion comment. A separate review traced both Azure list paths and reported 42 focused tests passing.
Check: GitHub check. Current-head Test passes. Check and Release Smoke also pass. Native analysis skips unchanged native files by workflow design.
Independent readiness check. The integrated change carries an optional byte cap through AzureDevOpsCli.execute into the existing VcsProcess and ProcessRunner limit. Only paginated PR-list requests use sixteen MiB. Other Azure commands retain the default cap. Tests cover a valid page above one MiB and verify propagation to the process layer.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head 5d46c7c6d669f0304cc80e4c821973d4e3f15498. The synthetic merged tree was not built or executed.
Scope limit. Pages larger than sixteen MiB can still fail decoding. This PR is a scoped cap increase, not unlimited collection.
Request. Stop rapidly failing managed connector processes from exhausting server resources.
Audit finding. Main still respawns an exited connector without delay. The complete two-file diff adds bounded exponential delay for rapid exits and tests stable-uptime reset and configuration preemption. Current-head required tests and checks pass with the native-only check skipped, the merge is clean, and the final review has no unresolved findings.
Recommendation. Keep open: work remains. Merge this connector restart-backoff fix.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Check: GitHub check. Current-head Test passes. Check and Release Smoke pass, with native static analysis skipped for this server-only change.
Comment: Discussion comment. Final approval covers the bounded restart, stable reset, and config-preemption behavior.
Pr: PR #8906. The alternative fixed-delay PR targets the same issue.
Independent readiness check. The integrated supervisor bounds rapid connector restarts while preserving an immediate first restart. The delay runs outside the reconcile lock. PID and configuration checks before and after the wait prevent a stale supervisor from replacing a newer connector. Stable uptime and explicit configuration reset the delay, and focused tests cover these paths.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head 76b84654d77b328db80001ed28bd270bca5aecbc. The synthetic merged tree was not built or executed.
Scope limit. Repeated crashes can delay recovery by up to sixty seconds. That tradeoff is explicit and avoids the unbounded spawn loop.
Request. Allow confirmed project removal when the only remaining threads are archived.
Audit finding. Project Settings sets force only when the active shell list contains threads, while the legacy sidebar has the same empty-project omission. The diff warns about archived history and sets force after confirmation on both paths. This fixes the web and desktop deletion request, but it does not change native mobile.
Recommendation. Keep open: work remains. Merge the confirmed archived-history removal fix.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Check: GitHub check. Current head has passing required checks, expected native-analysis skip for web-only changes, and approval.
Independent readiness check. The integrated change sets force only after destructive confirmation in Project Settings and both legacy-sidebar removal paths. The dialog now names archived history, which explains why an apparently empty project needs force. Backend deletion safeguards and environment/project routing are unchanged. Existing before/after dialog evidence covers both entry points.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head de578a3f058b8ab5cf47627fdb5d83b17820c15e. The synthetic merged tree was not built or executed.
Scope limit. Removing a project intentionally deletes its archived conversation history after confirmation. This does not remove project files on disk.
Audit finding. Main still stages npm installs without an allowScripts manifest or an explicit node-pty binary check. The diff adds both before publication, with focused success and failed-native-build coverage. The unchanged reviewed head is clean and approved, with Test, Check, and Release Smoke passing and native analysis correctly skipped; previously published incomplete runtimes still need repinning.
Recommendation. Keep open: work remains. Merge the reviewed fix and document repinning for an already-broken runtime.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Comment: Discussion comment. Latest-head approval; the separate Codex review reported no major issues.
Check: GitHub check. Latest reviewed head passed required Test, Check, and Release Smoke; native analysis was skipped by its path gate.
Independent readiness check. The integrated change stages an npm allowScripts manifest before install and checks the node-pty binary before validation, sentinel creation and atomic publication. Failure retains the existing staging cleanup path. The live npm 12 reproduction in the PR and the failed-native-build test cover the reported mechanism.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head d747c04c4dc8c32503c277cc08cf5b0daafda482. The synthetic merged tree was not built or executed.
Scope limit. A binary existence check is not an ABI test. The existing runtime validation remains in place.
Request. Stop session listing from scanning historical provider bindings.
Audit finding. Main still performs the historical binding scan, and this full diff limits reads to the active-session IDs without changing the returned session set. The focused test verifies that listThreadIds is not called and only the live thread binding is read. Test, Check, and Release Smoke pass on this head, Mobile Native Static Analysis is skipped for this server-only diff, and the review bots report no open findings. This is the preferred replacement for the earlier equivalent lookup PR.
Recommendation. Keep open: work remains. Merge this focused lookup fix and close the equivalent earlier PR as a duplicate.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Comment: Discussion comment. Current-head approval covers this narrow lookup change.
Check: GitHub check. Head CI: Test success, Check success, Mobile Native Static Analysis skipped, Release Smoke success.
Independent readiness check. The integrated method still returns only sessions from live adapters and applies the same persisted binding overrides and consistency checks. The patch restricts binding reads to the deduplicated thread IDs in that already-computed active set. Historical rows were never included in the returned list, so skipping their reads preserves output while bounding work by active sessions.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head a319843e007d1afcfb3c1e08cb0b36001aad67cc. The synthetic merged tree was not built or executed.
Scope limit. The number of reads still grows with active sessions. It no longer grows with total persisted thread history.
Request. Let quiet terminals persist and close while another terminal produces output continuously.
Audit finding. Main still recursively processes the same dirty key in the single worker, so another terminal can wait indefinitely. The patch atomically returns that key to the queue tail and has deterministic fairness and coalescing tests. Required CI and the current bot review pass on 0427b5f, and the package-relocation comment was withdrawn. This fixes scheduling fairness, not history growth or full-file rewrite costs.
Recommendation. Keep open: work remains. Merge the focused queue-fairness fix.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Check: GitHub check. Head 0427b5f: Check, Test, and Release Smoke pass. Mobile Native Static Analysis is skipped.
Independent readiness check. The integrated worker moves a dirty key from active to queued and offers it at the queue tail in one transaction. Pending data remains in latestByKey, so later enqueues still coalesce and drainKey still waits for all queued, active and pending work. Other keys can now run between batches. Existing failure requeue behavior is unchanged.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head 0427b5fbaa8ca4b0e76c9fdba47a135b6d00c982. The synthetic merged tree was not built or executed.
Scope limit. This fixes scheduling fairness only. It does not cap terminal history size or replace full-file persistence.
Request. Keep Start from origin visible even when new threads default to Local.
Audit finding. Main still hides the setting behind defaultThreadEnvMode === worktree, although a manually created worktree can use the preference. The one-file diff removes only that visibility guard. The existing preference, reset action, and worktree creation behavior stay unchanged.
Recommendation. Keep open: work remains. Merge the setting-visibility fix.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Check: GitHub check. Current head has passing required checks, expected native-analysis skip for web-only changes, and approval.
Independent readiness check. The integrated one-file change only removes the visibility condition around Start from origin. The checked value, setter and reset action are unchanged. The existing worktree decision uses the actual new thread mode, not the default mode, so manually created worktrees still use this setting when Local is the default.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head b8035413fc8b5dd8ecab400db0471574c7d84cc3. The synthetic merged tree was not built or executed.
Scope limit. The setting stays visible even when the default is Local. It still affects worktrees only.
Request. Center composer notice actions beside the complete title and description.
Audit finding. Main still places the description below the row that owns the actions. The full diff moves title and description into one content column and top-aligns the icon, with the prior review findings resolved. Current required checks pass and the head is approved; the broader padding PR should retain only its separate padding concern.
Recommendation. Keep open: work remains. Merge the alignment-only fix and keep the separate padding change under review.
Confidence high. Release: Not applicable. PR readiness: Safe to merge now.
Pr: PR #8890. Earlier active PR contains the same title-and-description alignment change.
Check: GitHub check. Current head has passing required checks, expected native-analysis skip for web-only changes, and approval.
Independent readiness check. The integrated layout puts title and description in the same content column. Current ComposerBanner.Row centers the separate action column against that block, while the new self-start icon rule preserves title alignment. The earlier transform workaround is gone, and both review findings are resolved on the audited head.
Merge simulation: clean. Audited main b883fc066ea5c9bebbe1c3e9b4bc2471aab3685f. Audited head 8a0e3475f54682d200f030053227261ff0164cb5. The synthetic merged tree was not built or executed.
Scope limit. All described standard banners share the new alignment, including Wake now, Compact and dismiss actions. Native mobile does not use this component.
Request. Use native Ctrl+Arrow and CSI Alt+Arrow sequences so Windows shells move by word instead of printing letters.
Audit finding. Main still intercepts non-macOS Ctrl+Arrow and Alt+Arrow and sends ESC b or ESC f. The PR removes the Ctrl override and changes Alt movement, including macOS Option movement, to CSI sequences, with focused mapping tests. No later keybinding change implements that shell compatibility change.
Recommendation. Keep open: work remains. Review the patch with a shell matrix for PowerShell, cmd, bash, and zsh.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. No terminal interaction test was run. Required checks absent from the canonical latest-commit rollup: Check, Test, Mobile Native Static Analysis, Release Smoke.
Request. Add a copy-path button to each diff file header.
Audit finding. Main exposes only the header-prefix slot in AnnotatableCodeView, and DiffPanel has no copy-path action. The complete diff adds the suffix slot and reuses clipboard feedback without changing source-control data. This remains a distinct UI addition, not a landed fix.
Recommendation. Keep open: work remains. Review the clipboard interaction in the current diff view after CI runs on the rebased head.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Keep branch selection in the current worktree when the selected ref has no checkout.
Audit finding. Main still clears activeWorktreePath when refName.isDefault is true and that ref has no existing worktree. The PR removes that special case while preserving reuse of an explicitly reported checkout. The exact behavior remains present on main.
Recommendation. Keep open: work remains. Review the default-branch checkout behavior and run the focused branch-selection tests on the current base.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Stop SSH host aliases beginning with a dash from being parsed as command options.
Audit finding. Both current command and tunnel argument arrays still place the user-controlled host directly after SSH options without an option terminator. The merged tunnel-process isolation change sets ControlMaster and ControlPath but does not protect the destination argument. This PR adds the separator at both spawn paths and tests its placement.
Recommendation. Keep open: work remains. Review and refresh the focused SSH argument change on current main.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Allow a custom keyboard shortcut to stop the active thread.
Audit finding. Main has no thread.stop command in the typed keybinding list. The patch routes an unassigned command to the existing interrupt request and checks running state before claiming the key event, so unrelated Escape actions remain available when idle. No provider protocol change is needed, but the settings command list and focused terminal or preview behavior need current-head validation.
Recommendation. Keep open: work remains. Rebase and verify a custom stop shortcut with the composer, terminal and preview focused.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. Required check did not execute on the recorded head: Mobile Native Static Analysis. GitHub reports conflicts with the target branch, so recorded checks do not validate a rebased integration.
Request. Stop timed-out background Git status fetches from filling the shared object store with temporary packs.
Audit finding. Main still fetches into the shared Git directory with a five-second timeout and has no owned temporary-pack cleanup. The final patch is much broader than its opening description: it fetches through private bare repositories, promotes objects locally, updates refs with compare-and-swap, and reaps old files. The fifteen review threads are resolved in the inspected diff, but unbounded uninterruptible promotion and age-based cleanup of shared legacy packs still need deliberate maintainer review.
Recommendation. Keep open: work remains. Verify slow multi-worktree fetch cleanup and review object promotion and legacy-pack ownership before merging.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Discover the Codex CLI bundled with Codex Desktop on Windows when PATH has no Codex executable.
Audit finding. CodexDriver still uses config.binaryPath directly for maintenance, probes and runtime creation. Main has no Codex Desktop runtime search, so the Windows fallback is still needed. The diff preserves explicit paths and PATH priority and adds focused resolver tests.
Recommendation. Keep open: work remains. Review the Windows fallback against current provider-instance setup.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Reject Cursor command wrappers that start the desktop editor during health checks.
Audit finding. Main still runs the configured command for the about probe without inspecting launcher wrappers. The default command fix and later opt-in default reduce accidental probes, but an enabled instance with a desktop-forwarding shim remains exposed. The PR adds a bounded file check before spawn and focused wrapper tests.
Recommendation. Keep open: work remains. Review the wrapper guard against the current Cursor probe and run its Windows cases.
Request. Discover favicons in the standard Svelte and SvelteKit static and src/lib/assets directories.
Audit finding. Main still omits both Svelte directory sets from FAVICON_CANDIDATES. Later favicon work improves scan performance, caching, and manual icon selection, but does not add these automatic lookup paths. The six added path cases and Windows-safe assertions cover the proposed server lookup change.
Recommendation. Keep open: work remains. Review this focused resolver patch after current CI completes.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Wait for remote Codex plugins before refreshing skills and opening threads.
Audit finding. The provider probe still requests skills/list immediately after initialization without plugin/installed or forceReload. Parent thread start and resume also have no plugin barrier. The PR addresses both entry points, but adds two sequential RPCs to session startup that need supported-version review.
Recommendation. Keep open: work remains. Review plugin/installed compatibility and the startup ordering tests on current main.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Select which connected environment supplies process, resource, trace, and log diagnostics.
Audit finding. Main still binds the diagnostics panel and its editor list to the primary environment. Query retry improvements cannot supply an environment in a client-only session or prevent wrong-machine pending action state. The diff adds a target selector, disconnected notices, and environment-keyed process signals, but needs review alongside the broader settings-target proposal.
Recommendation. Keep open: work remains. Review this focused diagnostics selector separately from the broader settings-target refactor.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. Current-head required checks were not reported: Check, Mobile Native Static Analysis, Release Smoke, Test. GitHub reports merge conflicts with main on the collected head.
Request. Remove hidden and internal-only Codex models from interactive model pickers.
Audit finding. parseCodexModelListResponse still maps every returned model without checking hidden or excluding codex-auto-review. The filtering diff addresses a current server catalog gap that affects all clients. No landed model-picker change applies this filter.
Recommendation. Keep open: work remains. Rebase and review the two-file catalog filter.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Back off checkpoint capture after repeated failures in the same workspace.
Audit finding. CheckpointStore on main still resolves the driver and attempts every requested capture without failure history. Existing checkpoint error typing does not cap repeated expensive failures. The PR adds a bounded workspace cache and cooldown while returning the same error, so the retry-control behavior is not landed.
Recommendation. Keep open: work remains. Review cooldown recovery and concurrent captures for one repeatedly failing workspace.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Show a bounded SQLite result code or schema issue in persistence error messages.
Audit finding. Main still hides the SQLite cause behind a repeated operation name. The final diff emits normalized SQLite codes or schema issue tags and fixes the earlier raw-message finding. The remaining review claim that SqlError.reason is unreachable is false for the installed Effect code, which exposes reason as cause, and all four required checks pass.
Recommendation. Keep open: work remains. Resolve the false-positive cause-chain review thread with the existing reproduction evidence.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Explain common Git clone authentication and network failures without sending raw stderr over RPC.
Audit finding. Current cloneRepository still lets Git command failures reach a generic source-control operation message and uses unknown for a raw URL provider. The PR classifies nonzero clone stderr locally and keeps a structured, bounded GitCommandError cause. Spawn failures and timeouts still use the existing generic wrapper, so this is not complete diagnosis for every clone failure.
Recommendation. Keep open: work remains. Review the common nonzero-exit classifications and add a timeout case to document the remaining path.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. No real SSH, HTTPS, or network failure was exercised. Required checks absent from the canonical latest-commit rollup: Check, Test, Mobile Native Static Analysis, Release Smoke.
Request. Render a duration that rounds to 60 seconds as 1m.
Audit finding. Main still renders every value from 59500 ms to just below 60000 ms as 60s, while its minute branch already rolls 60 seconds into the next minute. The PR changes only this shared formatting boundary and adds boundary and invalid-input tests. The shared function has no later fix in main history.
Recommendation. Keep open: work remains. Review the one-branch formatting fix with its focused tests.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Prevent a collapsed Android composer hint from wrapping into clipped text.
Audit finding. The current Android setter only changes gravity and still permits multiple lines. This three-line change limits collapsed text and restores multiline editing on expansion. It does not change iOS or web editing.
Recommendation. Keep open: work remains. Refresh native checks and review the narrow Android layout change.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Check: PR #5052. Current head b57e6166: required checks are absent from canonical latest-commit metadata: Test, Check, Mobile Native Static Analysis, Release Smoke.
Limits. Required CI checks are not present on the current head in canonical check metadata.
Request. Emit a deletion tombstone when a client asks to delete an already-missing thread.
Audit finding. The main decider still calls requireThread before emitting thread.deleted, so missing threads fail instead of clearing stale client entries. The merged missing-worktree fix handles filesystem cleanup, not this missing aggregate case. The focused patch remains relevant, but reliable detail-subscriber cleanup also depends on the separate deletion-event fix.
Recommendation. Keep open: work remains. Review the idempotent decider change and pair its stale-client test with deletion-event delivery.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Render workspace HTML in the file panel for web users as well as desktop users.
Audit finding. Main still limits the Source/Rendered toggle to Markdown and the browser-preview button to supported desktop runtimes. The PR adds a sandboxed srcdoc iframe with signed relative-asset URLs and keeps source as the default. The newer media PR overlaps HTML delivery but has not supplied this behavior on main.
Recommendation. Keep open: work remains. Review the HTML iframe patch together with the media-delivery branch and verify relative assets over a remote connection.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Pr: PR #9023. Open cross-client media work overlaps asset delivery.
Limits. The sandbox and remote asset paths were reviewed in source, not in a browser. The collected latest head has no results for these required checks: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Allow a preview URL for an absolute file outside the project directory.
Audit finding. Main still converts absolute paths relative to the workspace and rejects paths outside that root. The complete diff issues an exact-file capability for an external file while preserving workspace sibling assets and relative-traversal rejection. The larger media branch addresses this too, but it remains open.
Recommendation. Keep open: work remains. Choose whether this bounded server fix should land separately from the larger media branch.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Keep the final iOS file rows above the floating search toolbar.
Audit finding. Main still gives FileTreeBrowser eight points of bottom padding while the iOS screen overlays a 112-point toolbar fade. The PR adds route-specific padding and leaves Android and the iPad navigator unchanged. The source mismatch remains, but an iOS runtime check is still missing.
Recommendation. Keep open: work remains. Verify the last file row on iOS with enough files to fill the list.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Mobile network changes and reconnects should quickly probe sockets and renew the server activity lease.
Audit finding. Main has application-resume wakeups, but no network-path-changed wakeup and no connected-generation trigger in the activity reporter. The current PR keeps those distinct gaps after removing earlier reconnect changes already covered by main. Its 30-second idle timeout only affects Bun servers, so Node behavior and actual WiFi-to-cellular recovery still need verification.
Recommendation. Keep open: work remains. Verify a mobile network switch and reconnect lease renewal against both Node and Bun before reviewing the timeout change.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Advertise each usable desktop IPv4 interface as a distinct selectable pairing endpoint.
Audit finding. Main still returns the first usable IPv4 address from networkInterfaces, and all desktop LAN endpoints share one stored preference key. The merged QR endpoint picker lets a user choose an advertised endpoint but cannot expose the unlisted interfaces. This diff adds enumeration, per-address keys, live interface refresh, and Tailscale deduplication.
Recommendation. Keep open: work remains. Review the multi-interface endpoint list and stored-default fallback with VPN and LAN adapters.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Merged pr: PR #5360. Adds endpoint choice to QR sharing, not discovery of every interface.
Limits. Current-head required checks were not reported: Check, Mobile Native Static Analysis, Release Smoke, Test. GitHub reports merge conflicts with main on the collected head.
Request. Prevent cached provider snapshots from restoring deleted custom models.
Audit finding. Cache hydration still appends every absent cached model, and live refresh still retains absent prior models without excluding isCustom entries. The patch separates configuration-backed models from discovered models in both paths and keeps discovered-model fallback behavior. The immediate composer problem is separate and still has its own open patch.
Recommendation. Keep open: work remains. Review the cache filtering and run startup plus incomplete-refresh tests with two provider instances.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Make native mobile bare URLs with md4c-rejected underscores tappable.
Audit finding. Main still sends markdown straight to parseMarkdownWithOptions without the proposed text-node autolink pass. The patch covers the iOS selectable renderer, Android fallback messages, and file previews, with tests for opaque code nodes and balanced parentheses. This is parsing behavior, so the later safe URL-opening change does not replace it.
Recommendation. Keep open: work remains. Refresh the AST transform and verify the reported URL through the current iOS renderer.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Allow wheel and trackpad scrolling through overflowing right-panel tabs on desktop.
Audit finding. Main still marks the ScrollArea itself as an Electron drag region, so the proposed no-drag change is absent. The final patch keeps only an empty trailing spacer draggable and cancels its extra gap, addressing the review findings about resize handles and false overflow. It is desktop-specific behavior, although the same component serves web tabs.
Recommendation. Keep open: work remains. Run a native desktop check of overflow scrolling, title-bar dragging and panel resize on the final patch.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Refresh and order network-status updates consistently when a client resumes.
Audit finding. Main already re-reads mobile network status on AppState active, so the basic missed-listener recovery landed with the mobile-resume change. It does not serialize those asynchronous snapshots against later network reports, and relay discovery still handles online/offline changes separately. The PR adds shared ordering and discovery-state fixes that are not covered by query retries or the existing mobile adapter.
Recommendation. Keep open: partial fix. Review the shared ordering change against live-report races and relay discovery without duplicating platform refreshes.
Request. Draw solid Powerline separators across the full terminal cell.
Audit finding. The Ghostty canvas renderer still treats the two solid Powerline characters as font text. The patch draws cell-sized triangles, separates them from text runs and handles block-cursor repaint. The maintainer requested before-and-after evidence and the author supplied it in discussion, but current required checks are not available.
Recommendation. Keep open: work remains. Rebase and verify both separator directions under selection, inverse colors and block cursors.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Match equivalent shortcut spellings when replacing or removing saved keybinding rules.
Audit finding. Main still compares raw key strings, while Settings renders escape as esc, so replacing the saved rule leaves the old rule behind. The earlier duplicate-rule fix does not normalize aliases. The unresolved literal-space review finding is outside validated input: KeybindingValue trims the string and rejects an empty key, as the author explains.
Recommendation. Keep open: work remains. Review the alias normalization and the focused persisted-rule test.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Keep usage refresh pending until each environment finishes its current request.
Audit finding. Main derives answeredCount from the presence of any summary, including a retained stale value, rather than the current waiting flag. The patch separates connection state, pending refresh and terminal failure and covers reconnects without changing the wait-for-all presentation choice. The later usage redesign does not fix this shared state calculation.
Recommendation. Keep open: work remains. Port the refresh-generation status helper to the current usage page and test a slow second environment.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. WSL backend discovery should use the route-selected address instead of the first hostname address when Docker adds interfaces.
Audit finding. Main still chooses the first IPv4-looking hostname -I token, so Docker bridge order can still select the wrong address. The PR asks the kernel for a route source and accepts hostname fallback only when unambiguous. The later WSL runtime-cache change affects startup files, not address selection, and mirrored WSL with Docker remains unverified.
Recommendation. Keep open: work remains. Run the route-source change on the reported mirrored-WSL and Docker configuration.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Check: PR #5889. Head 34a6876. Required checks absent: Check, Test, Mobile Native Static Analysis, Release Smoke. GitHub reports conflicts with main.
Limits. The reported mirrored-WSL plus Docker topology was not reproduced. Required CI is absent on the current head: Check, Test, Mobile Native Static Analysis, Release Smoke.
Request. Change balanced-profile remote Git polling from thirty seconds to five minutes.
Audit finding. Main still defines the balanced interval as thirty seconds. The recent idle CPU change reduces work inside Git status refreshes but does not change this default. The PR preserves performance mode and explicit overrides, so the remaining question is the acceptable delay for remote branch and pull-request status.
Recommendation. Keep open: decision needed. Decide whether five-minute remote status freshness is the intended balanced default.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Limit the thread action context menu to the thread title.
Audit finding. Main still attaches handleHeaderContextMenu to the full flex-width header and excludes only the action area. Double-click rename was added later, but it does not narrow that right-click target. The small patch remains useful and must preserve the newer title double-click handler.
Recommendation. Keep open: work remains. Rebase the title-only context-menu change while retaining double-click rename.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. Canonical current-head results are missing for required checks: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports conflicts with the target branch, so recorded checks do not validate a rebased integration.
Request. Recover desktop startup when the outer connection-catalog JSON document is corrupt.
Audit finding. Main still propagates readDocument decode failures before it can run missing-catalog recovery or legacy migration. The existing writer already uses a temporary sibling and rename, so atomic-write history does not recover an already malformed document. The PR limits recovery to the outer document decode error and leaves read, encryption, and decryption failures visible.
Recommendation. Keep open: work remains. Review the quarantine fallback with the existing catalog corruption and secure-storage tests.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Render available usage totals without waiting indefinitely for an unreachable environment.
Audit finding. Main still holds the web usage page while isPending or isPartial and gives usageSummary no per-environment deadline. The merged query retry work restores interrupted requests but cannot make an unreachable environment answer or release this all-environments display gate. This PR adds partial totals and a bounded query wrapper shared with mobile, which needs review against the current retry semantics.
Recommendation. Keep open: work remains. Rebase the bounded usage query and verify one reachable plus one permanently unreachable environment.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Add separate keyboard commands to move focus between the terminal and composer without hiding the terminal.
Audit finding. Main has terminal.toggle but does not register terminal.focus or chat.focusComposer. The PR adds the commands, opposite focus-context defaults, and separate focus requests for the drawer and right-panel terminal. Current visibility-toggle behavior therefore does not cover this requested keyboard route.
Recommendation. Keep open: work remains. Review the focus-target behavior with both terminal hosts visible and a maximized right panel.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Offer last-modified ordering in the Add Project folder browser.
Audit finding. Main still sorts directory entries by name and has no sort-order input or client preference. The patch keeps name ordering as the cheap default and only stats visible directories with bounded concurrency when modified ordering is selected. Native mobile intentionally keeps name ordering, while web and desktop send the new optional field.
Recommendation. Keep open: work remains. Review the optional folder-order setting and run the directory-order tests on current main.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. Canonical current-head results are missing for required checks: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports conflicts with the target branch, so recorded checks do not validate a rebased integration.
Request. Windows desktop titlebar geometry briefly shifts when a minimized window is restored.
Audit finding. Main still removes the wco class whenever the overlay reports hidden, with no saved valid geometry or Windows fullscreen synchronization. The PR preserves the control inset during restore and updates several headers, while recent workspace navigation changes do not add that stabilization. The first visible event intentionally reuses old geometry, so restore and DPI changes need native verification.
Recommendation. Keep open: work remains. Rebase the geometry change and verify minimize, restore, fullscreen, and DPI changes on Windows.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. The empty Electron workspace needs a draggable titlebar when the sidebar is collapsed.
Audit finding. NoProjectsHero on main still starts directly with its empty-state content and has no drag-region strip. The PR adds a small Electron-only spacer and reports a native Windows drag with the sidebar collapsed. Its static-markup tests only assert attributes, so the reported native capture is the meaningful behavior evidence.
Recommendation. Keep open: work remains. Review the spacer against the current empty workspace and retain native Windows drag evidence.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Show a useful binary-file state instead of a raw file-read error.
Audit finding. The query layer still reduces failures to message text, and FilePreviewPanel still displays that text for binary reads. The new reveal-file action does not add a binary-preview state. The PR preserves the typed failure and offers Copy path, leaving other errors unchanged.
Recommendation. Keep open: work remains. Review the binary-file state using the current query and file-preview components.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Keep Windows resize cursors visible in light themes.
Audit finding. The right-panel handle and body drag cursor still use col-resize, and both terminal resize handles still use row-resize. Theme changes did not replace these Chromium cursor types. The full diff changes these controls to the system ew-resize and ns-resize cursors without changing resize behavior.
Recommendation. Keep open: work remains. Review the cursor-only change with its Windows evidence.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Let users choose an exact snooze date and time on web, desktop, iOS, and Android.
Audit finding. Main accepts a future snoozedUntil value but exposes only presets, so recent preset and server-settling fixes do not add this feature. The full diff covers row, context-menu, chat, bulk, and adaptive mobile entry points while reusing existing snooze and wake commands. The reported Android DST and empty closing-dialog findings are fixed in the final diff, but the patch must preserve the newer server-owned settling behavior when rebased.
Recommendation. Keep open: work remains. Review the custom-time feature on current main with DST, bulk failure, Undo, and Wake checks.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Comment: Discussion comment. Final Android implementation preserves wall-time fields and tests DST gaps and repeats.
Limits. The reported desktop and native-device verification was not repeated here. Latest-head Mobile Native Static Analysis was skipped, not executed.
Request. Keep an explicit healthy source-control writer when stored provider settings lag behind its live state.
Audit finding. The current resolver rejects an explicit writer using stored provider settings before it checks the live provider snapshot. The PR reverses that precedence when a snapshot exists and adds Create PR coverage for OpenCode model options. This specific mismatch is still present on main.
Recommendation. Keep open: work remains. Review the live-provider precedence and run the Create PR selection test.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Reduce Windows editor discovery probes by listing PATH directories before checking executable candidates.
Audit finding. Main still probes each PATH and PATHEXT candidate and caches only per-command results. The diff reduces first-pass probes on Windows while retaining direct probing when directory listing fails. The directory-cache PR targets the same bottleneck with broader reuse, so the two implementations need one owner.
Recommendation. Keep open: work remains. Choose this Windows-only resolver change or the shared directory-cache implementation before merging either.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Audit finding. All four snapshot SELECT paths and the session repository still expose active_turn_id without blank-to-null normalization. The patch adds normalization at each read boundary and checks both active and archived threads while preserving valid IDs. Canonical current-head checks pass, but the branch conflicts with current main and still needs a rebased projection check.
Recommendation. Keep open: work remains. Run the focused projection tests on a rebased patch with empty, space-only and valid IDs.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Recognize credentialed HTTPS and explicit-port SSH GitHub remotes in repository parsing.
Audit finding. The shared parser and its GitManager copy still reject these two URL forms on main. The SSH-user fix changed provider detection, not these owner/repository parsers. This PR updates the regex and removes the duplicate parser, so the parsing gap remains open.
Recommendation. Keep open: work remains. Review the parser update with the credentialed HTTPS and explicit SSH-port tests.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Render completed Mermaid fences as diagrams across the shared web Markdown renderer.
Audit finding. Main still renders Mermaid through its ordinary code-block path and has no Mermaid runtime dependency. This patch lazy-loads strict-mode rendering, skips obsolete queued renders and preserves code-copy behavior and intrinsic diagram width. Another Mermaid implementation remains open, so neither is a landed replacement; native mobile rendering is outside this patch.
Recommendation. Keep open: decision needed. Choose one Mermaid implementation and confirm the shared-renderer and mobile scope before merging.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Pr: PR #8533. Separate open Mermaid-rendering proposal.
Limits. Canonical current-head results are missing for required checks: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports conflicts with the target branch, so recorded checks do not validate a rebased integration.
Request. Remote Update and Retry actions must wait until their environment connection is ready.
Audit finding. Main still renders an enabled self-update button and guards its handler only against an update already pending. The PR adds a reactive connected-phase check to both the button and handler while leaving the copy-command path available offline. Later server-update work has not added this guard.
Recommendation. Keep open: work remains. Review and land the connection guard with the offline-to-connected interaction test.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Start on the project checkout when a requested worktree has no repository or base commit.
Audit finding. Merged PR 8226 allows a deleted draft ID to be reused after a failed bootstrap, and PR 7664 added client retry handling. Main still creates the thread before worktree preparation and has no repository or base-commit fallback. The duplicate-ID symptom improved, but non-Git and unborn-HEAD worktree requests still need the proposed preflight.
Recommendation. Keep open: partial fix. Review a preflight before thread.create and preserve the existing committed-base setup path.
Request. Prevent Windows editor discovery timeouts with a shared PATH-directory listing cache.
Audit finding. Main still caches complete command lookups and performs per-candidate filesystem probes on a cold scan. This diff shares directory listings across commands and preserves fallback probing for unreadable directories, which is not provided by recent PATH ordering fixes. It overlaps the narrower Windows-only listing PR and changes lookup caching on all platforms.
Recommendation. Keep open: work remains. Select the directory-cache approach and verify macOS and Linux executable resolution before merging.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Pr: PR #6124. Narrower Windows-only alternative for the same cold-discovery cost.
Limits. Reported live verification covers Windows only. Required CI has no result on the reviewed head: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Attach annotation screenshots without a data-URL fetch blocked by desktop CSP.
Audit finding. Main still fetches annotation.screenshot.dataUrl while the desktop connect-src list excludes data URLs. PreviewView also keeps the screenshot claim when conversion fails. The complete patch decodes locally and clears that claim on failure, so both parts still have work to do.
Recommendation. Keep open: work remains. Review the local-decode fix and verify draw, region, and element annotations in desktop.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Include the resolved component file in desktop element picks even when the call stack is empty.
Audit finding. Main still uses stack[0] as the only source frame and labels the location as source. The complete diff adds the independent react-grab location with stack fallback and clarifies the component-definition label. Existing browser input changes did not change this capture path.
Recommendation. Keep open: work remains. Review the source mapping against an element whose resolved file exists but stack is empty.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Upgrade the Homebrew formula variant that actually owns a provider binary.
Audit finding. Main still constructs brew upgrade with the driver's fixed homebrewFormula. The diff resolves Cellar and Caskroom package names, keeps matching tap-qualified defaults, and adds tests for variant formula names. The recent npm update fix does not address Homebrew variants.
Recommendation. Keep open: work remains. Run current provider-maintenance checks and obtain maintainer review for the formula-resolution change.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Keep Merge unavailable until the post-merge detail refresh settles.
Audit finding. Main clears pendingAction immediately after the action RPC and then refreshes detail, leaving a stale-open interval. The patch adds per-PR state and limits the hold to Merge, with an escape when the next detail read finishes even if the host still says open. This is not covered by server-side thread settlement and still needs current-head verification for GitHub and asynchronous Azure completion.
Recommendation. Keep open: work remains. Verify the per-PR hold with delayed and failed detail reads and switching between two PRs.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Do not send a kill request to a provider updater that has already exited.
Audit finding. The maintenance runner still installs an unconditional child.kill finalizer. The npm update fix changes install-script permission, not this cleanup path. This PR adds an isRunning check and a test for an exited child, so the redundant Windows taskkill path remains unfixed on main.
Recommendation. Keep open: work remains. Review and run the maintenance runner test on the latest branch.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Enable Git long-path support per command on Windows without writing user configuration.
Audit finding. Main still passes inherited environment values to Git without injecting core.longpaths. The longer worktree-add timeout and missing-worktree cleanup do not change Git for Windows path handling. This PR preserves inherited GIT_CONFIG entries and adds Windows-only injection, so the stated creation and removal gap remains.
Recommendation. Keep open: work remains. Review inherited configuration handling and verify create and remove with a Windows path longer than 260 characters.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Keep the traits menu open while selecting multiple model options.
Audit finding. Both select and boolean trait rows on main still set closeOnClick. The patch changes only those dismissal flags, so it covers web and desktop without changing provider options or remote transport. This is a menu behavior choice, not a defect already fixed by later model work.
Recommendation. Keep open: decision needed. Decide whether trait selection should keep the menu open, then verify Escape and outside-click dismissal.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Drag right-panel tabs into a persistent custom order.
Audit finding. Main exposes no reorderSurface action and still groups browser tabs during reconciliation. The patch persists arbitrary ordering, activates a grabbed tab and suppresses active-tab scrolling during the drag. It covers chat and pull-request tab strips, but must be rebased without dropping newer browser audio controls or the pending desktop scroll fix.
Recommendation. Keep open: work remains. Rebase the reorder behavior while preserving current tab controls and verify overflow dragging on desktop.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Rename the focused sidebar thread with configurable F2 and return focus to its row.
Audit finding. Main still has no thread.rename keybinding, despite supporting pointer rename. The patch covers both sidebars and the settings recorder and now limits the autofocus exception to keyboard-visible focus. Its final diff addresses the reported Space activation, delayed focus theft and stale shortcut-context problems, but changes enough focus behavior to need an integrated keyboard pass.
Recommendation. Keep open: work remains. Verify F2, Enter, Escape, Space and pointer-to-composer focus in both sidebar modes on current main.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Quote selected assistant text into a new composer reply.
Audit finding. Main has no selected-text Reply action in AssistantTimelineRow. The patch restricts selection to rendered Markdown, refreshes the quote when selection changes and fixes release-on-padding behavior raised in review. It intentionally adds quote creation only to web and desktop, with native mobile limited to displaying the resulting Markdown.
Recommendation. Keep open: decision needed. Decide whether to ship web-only quote creation, then verify keyboard selection and floating-action dismissal.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. Canonical current-head results are missing for required checks: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports conflicts with the target branch, so recorded checks do not validate a rebased integration.
Request. Avoid showing identical raw-command and detail text twice in expanded tool rows.
Audit finding. Main still appends raw command and detail as separate blocks without checking equality. The patch deduplicates only identical trimmed blocks and retains distinct output, with focused tests for both cases. Later work-log presentation changes have not removed this code path.
Recommendation. Keep open: work remains. Rebase the display helper extraction and run its duplicate-command and distinct-output tests.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. Canonical current-head results are missing for required checks: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports conflicts with the target branch, so recorded checks do not validate a rebased integration.
Request. Let Android split-view users hide and restore the thread sidebar.
Audit finding. The Android thread and file-tree headers still have no sidebar toggle, while the iOS toolbar has one. WorkspaceEmptyDetail still assumes the sidebar is visible and offers only New Task. The patch covers collapse and Home recovery but intentionally leaves the Android file-viewer header without its own toggle.
Recommendation. Keep open: work remains. Refresh the Android split-view controls and verify collapse, back, and reveal on a foldable.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Fetch GitLab merge-request heads from their own ref namespace and avoid nonexistent refs on other hosts.
Audit finding. Main still hardcodes refs/pull at both branch and commit fetch sites, so the GitLab worktree failure remains after the newer worktree and push fixes. The full patch changes both initial materialization and reused-worktree refresh, keeps GitHub behavior, and handles no-head-ref hosts without confusing same-named fork branches. The reviewed buried-error defect is fixed in the final diff, with local Git integration coverage.
Recommendation. Keep open: work remains. Review the host-specific ref routing with current GitManager tests.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Show GitHub label descriptions in the pull request Summary tooltip.
Audit finding. Main labels still contain only name and color, and the GitHub mapper drops descriptions. The final patch uses the shared Tooltip, fixing the reviewed Popover mismatch, and includes descriptions only in detail responses to avoid expanding list traffic. Other source hosts and the native mobile UI gain no description display in this patch.
Recommendation. Keep open: work remains. Review the optional detail-only field and confirm the intended GitHub web and desktop scope.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Restore the desktop-width sidebar open state after refresh.
Audit finding. SidebarProvider still initializes from defaultOpen and writes a cookie that it never reads. The shelf-collapse persistence merge stores a different preference and does not handle the main sidebar. The localStorage read and write in this patch are still needed.
Recommendation. Keep open: work remains. Review the localStorage persistence change and verify both expanded and collapsed refresh states.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Show a count of unread background threads on the main sidebar toggle.
Audit finding. SidebarTrigger on main still renders only its icon and accessible label. This patch counts completed non-archived background threads with the existing local visit timestamps, so the separate server unread-state proposal would need to be reconciled if both land. It covers desktop and narrow web layouts, not the native mobile client.
Recommendation. Keep open: work remains. Review the badge behavior and coordinate its unread source with PR 6662.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Show live context-window usage for OpenCode sessions.
Audit finding. Main emits no OpenCode thread.token-usage.updated events. This PR derives a meter from assistant token counts and reads model limits plus compaction settings, without changing historical Usage-page collection. It overlaps the other live-meter proposals and waits for metadata on the sequential event path, so token semantics and the two-second event delay need a decision before one implementation is selected.
Recommendation. Keep open: work remains. Choose one live-meter implementation and verify token totals, compaction, and non-blocking metadata probes.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. macOS editor discovery should find bundle-only Zed and not mistake the standalone agy CLI for Antigravity IDE.
Audit finding. Main still discovers both editors only through their command names, and its editor definitions have no app-bundle metadata. The PR checks both Applications directories, requires the Antigravity IDE bundle, and launches bundle-only editors with open. Position suffixes are deliberately removed for bundle launches, so line navigation is not preserved on that fallback.
Recommendation. Keep open: work remains. Review the bundle fallback with Zed and Antigravity installed in each supported Applications directory.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Copy the complete persistent thread error and report clipboard success or failure correctly.
Audit finding. The current error banner exposes its full text in a tooltip but has only a dismiss button. The earlier dismissal fix did not add copying, and useCopyToClipboard still accepts stale asynchronous completions. The patch covers web and desktop while preserving session dismissal.
Recommendation. Keep open: work remains. Review the shared clipboard-hook changes and merge the banner control after current checks pass.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Include archived Codex rollouts in usage totals without counting moved files twice.
Audit finding. UsageService still scans only the Codex sessions directory even though the home layout shares archived_sessions too. This PR adds the second root and claims a rollout basename only after a nonempty read, covering the move-between-scans case. The new append reader changes the integration point, not the missing archived-root behavior.
Recommendation. Keep open: work remains. Rebase the archived-root scan and preserve its moved-rollout test on the current reader.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Make preview_evaluate return a serializable value even when returnByValue is false.
Audit finding. Main still forwards input.returnByValue to CDP and then JSON-encodes only result.value. A false flag can therefore yield an unusable object handle and fail encoding. The full PR forces by-value evaluation, retains argument compatibility, and adds a focused fake-CDP test.
Recommendation. Keep open: work remains. Review and rerun the focused evaluate test on current desktop main.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Open View PR in the external browser when Command or Control is pressed.
Audit finding. Main discards the click event in both the quick action and dropdown, and openExistingPr always prefers the in-app callback. The PR passes modifier state through both entry points without changing plain clicks. This remains separate from modifier support on actual PR links.
Recommendation. Keep open: work remains. Review both View PR entry points with plain click and Command or Control click.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Enable environment activity publishing when a push-capable phone links it.
Audit finding. Main still ends mobile linking after relayConfig and the server still defaults publishAgentActivity to false. This leaves notification-only iOS users silent as well as Live Activity users. The patch addresses both cases, unlike the Live-Activity-gated alternative, but changes environment preferences during linking and makes that request part of link success. The alleged relay:write blocker is a false positive because linkProof and relayConfig already require the same scope with the same token.
Recommendation. Keep open: work remains. Review the publish opt-in behavior and reconcile it with PR 4998.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Source: apps/server/src/cloud/http.ts:818. Preferences uses the same relay-write scope as the earlier link calls at lines 428 and 500.
Check: PR #7106. Current head 3528478b: required checks are absent from canonical latest-commit metadata: Test, Check, Mobile Native Static Analysis, Release Smoke.
Limits. Required CI checks are not present on the current head in canonical check metadata.
Request. Keep the native Electron User-Agent to avoid a Cloudflare Turnstile challenge loop.
Audit finding. Main still removes Electron and app tokens from the preview session User-Agent. The PR reports a controlled Electron 41 experiment, but main has since moved to Electron 43.4.1 and added native OAuth popups. The small diff is clear, but the current Turnstile result and embedded OAuth tradeoff need fresh verification.
Recommendation. Keep open: retest. Repeat the native-versus-rewritten User-Agent check on Electron 43.4.1 and include the current OAuth popup flow.
Confidence medium. Release: Not applicable. PR readiness: Needs maintainer review.
Commit: Commit 4a9d2d0ced2a. Main has moved to Electron 43.4.1 since the reported experiment.
Limits. The reported challenge experiment used Electron 41 and was not repeated on current main. The collected latest head has no results for these required checks: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Make unsent draft sessions reachable from the legacy sidebar.
Audit finding. Main renders SidebarDraftBlock only in the default sidebar. LegacySidebar still starts its content with the Projects section and provides no saved-draft list, even though the draft store retains those sessions. The draft-list merge for the default sidebar therefore covers only the other sidebar mode.
Recommendation. Keep open: work remains. Review the shared draft block in the legacy sidebar with existing and empty draft lists.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Commit mobile font-size changes once on drag release instead of on each gesture update.
Audit finding. Main still calls runOnJS(commit) during every pan update, which changes app-wide typography while dragging. The diff keeps motion in shared values, commits once after a successful gesture, and restores the old value on cancellation while preserving tap and accessibility actions. The Android before-and-after evidence supports the change, and the recent theme refactor did not implement it.
Recommendation. Keep open: work remains. Run current focused mobile checks and review the release-only commit behavior on iOS.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. The body reports Android interaction checks, not an iOS pass. Required CI has no result on the reviewed head: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Find and launch the macOS VS Code app when its code command is not on PATH.
Audit finding. Main still checks each editor static command list and does not add the standard VS Code bundle paths. The PR extends the same candidate list for discovery and launch, keeps PATH commands first, and tests an empty-PATH bundle installation. The broader app-bundle proposal is related work, not a landed replacement.
Recommendation. Keep open: work remains. Review the narrow bundle-path fallback with current launcher tests.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. No macOS application launch was performed. Required checks absent from the canonical latest-commit rollup: Check, Test, Mobile Native Static Analysis, Release Smoke.
Request. Share concurrent explicit VCS status refreshes for one checkout without cancelling work when one caller leaves.
Audit finding. Main still invalidates the workflow cache and starts local and remote reads on every explicit refresh. The proposal adds per-checkout shared fibers and generation-based reruns after mutations, with gated concurrency tests. Recent idle polling changes do not add this sharing.
Recommendation. Keep open: work remains. Review the invalidation and rerun races on current main, then run the focused broadcaster tests.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. JetBrains file links need the known workspace root so the editor opens the correct project window.
Audit finding. Current line-column launches still send only the file position and file path, and LaunchEditorInput has no projectRoot field. The PR carries the known root through the main web launch paths and changes only positioned JetBrains calls. Its tests cover argument shapes, but they do not prove which real JetBrains window receives a file.
Recommendation. Keep open: work remains. Verify a positioned file open with two JetBrains worktrees and then review the optional-root contract change.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Keep newly created SQLite database families local to a Codex shadow home.
Audit finding. Main still considers each shared-home entry separately and attempts to symlink new SQLite files. A regular shadow copy therefore reaches the existing conflict path, with no base/WAL/SHM family rule. The PR adds that rule without deleting regular files or relaxing unrelated config conflicts.
Recommendation. Keep open: work remains. Review the SQLite-family ownership change with the other shadow-home changes.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Let a project point to its renamed workspace folder through a relink command.
Audit finding. The current project CLI exposes add, remove, and rename, but no relink command. Its path lookup also requires normalizing an existing directory, so the old missing path cannot identify the project. This patch adds an explicit repair command rather than automatic folder-move detection, and it deliberately leaves existing sessions at their old working directory.
Recommendation. Keep open: work remains. Review the relink command as a manual repair and document its effect on existing sessions.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Cycle a model's advertised reasoning levels with configurable keyboard shortcuts.
Audit finding. Main still lacks reasoning.increase, reasoning.decrease, and the shared reasoning transition. The latest patch addresses the reported slash-command, caret, pending-answer, and held-key defects, with current required checks passing. Its mod+alt defaults and shared prompt-prefix behavior now need maintainer review, not another fix for the earlier reports.
Recommendation. Keep open: work remains. Review the default shortcuts and shared prompt-prefix behavior on the latest head.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Use a completed assistant message containing a change-request link as a remote-status refresh hint.
Audit finding. Main handles message-sent in CheckpointReactor only to establish a pre-turn baseline. It does not refresh remote status when an assistant posts a PR or MR link. The proposal covers four source-control URL forms and ignores streaming chunks, but adds network work to the checkpoint event worker.
Recommendation. Keep open: work remains. Review whether this refresh belongs in the checkpoint worker and confirm it does not delay checkpoint capture.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Keep desktop and web sessions connected through one slow foreground health probe before a bounded retry.
Audit finding. The merged stall fix raised desktop probe tolerance to 15 seconds, but main still replaces the lease after the first timeout. The query-retry fix only retries interrupted queries and does not change this connection decision. This PR adds one bounded desktop/web retry while leaving the mobile three-second probe unchanged.
Recommendation. Keep open: partial fix. Review the bounded desktop probe retry against the existing stall tests.
Request. Treat Cursor API-key-only installations as authenticated without opening a browser login.
Audit finding. Main derives Cursor authentication from agent about and always sends cursor_login during ACP setup. It has no CURSOR_API_KEY override on these paths. The PR changes both status and session startup, so the defect is not covered by the Cursor default-command or opt-in fixes.
Recommendation. Keep open: work remains. Verify the API-key probe and real ACP startup for a key-only Cursor instance.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Keep the Windows cross-architecture payload test valid when Electron environment variables are inherited.
Audit finding. The cross-architecture test still identifies the primary probe only by ELECTRON_RUN_AS_NODE. Recent packaging changes moved the test but retained the same assertion, so ordinary Node probes can still be mistaken for the packaged executable. The patch narrows only the test predicate.
Recommendation. Keep open: work remains. Review the executable-path assertion and run this focused suite from an Electron-parent environment.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Restart mobile connection setup after an iOS Local Network permission prompt returns the app to active state.
Audit finding. Main records only background transitions and interrupts establishment only for application-active-reconnect. The merged mobile-resume work added that long-suspension path but still ignores a short inactive-to-active probe during setup. The PR changes that exact gap and retains the connected-phase probe behavior.
Recommendation. Keep open: work remains. Verify first-time iOS LAN pairing with the Local Network permission prompt on the updated branch.
Request. Show the full server version on each saved remote connection and mark cached offline versions as last seen.
Audit finding. Current saved-connection metadata only shows the SSH target and T3 Connect label. Exact versions still appear only in the mismatch notice, so matching-version and cached offline rows lack the requested information. The diff reuses existing serverConfig state and adds no network request.
Recommendation. Keep open: work remains. Review the connected and cached-offline labels in web and desktop Connections settings.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. Current-head required checks were not reported: Check, Mobile Native Static Analysis, Release Smoke, Test. GitHub reports merge conflicts with main on the collected head.
Request. Move the unchanged Node SQLite Effect client to shared code for desktop reuse.
Audit finding. Main still keeps NodeSqliteClient under server persistence and loads it by relative import. The complete patch is a 100-percent file move with import and export updates, not a database behavior fix. It remains stacked on the unmerged browser-profiles branch; current Test, Check, and Release Smoke checks pass, while native analysis is skipped, so the parent dependency still prevents a standalone safe-merge recommendation.
Recommendation. Keep open: work remains. Retarget the mechanical move to main or land the browser-profiles dependency first.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Observed GitHub metadata. GitHub state OPEN. Author juliusmarminge. Updated 2026-08-29T07:55:29Z. Draft no. Target browser-profiles. Head 5c239bf34f0ea612c6b38bca6626dc61200e2e8a. Branch shared-sqlite-client. Size +22 / -18, 20 files, 2 commits. Mergeability MERGEABLE. Merge state CLEAN. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Request. Launch Cursor with its full-access flag so the CLI does not raise tool approval prompts.
Audit finding. Main still launches Cursor ACP without force and does not pass runtimeMode into its spawn builder. It auto-answers permission requests when an allow option exists, but this does not change the CLI allowlist behavior. The PR adds the flag only for Full access and records the remaining Auto behavior in user docs.
Recommendation. Keep open: work remains. Verify force with the supported Cursor CLI and the existing permission fallback.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Report each Claude subagent model and effort without guessing from its parent.
Audit finding. The merged model-race fix already buffers early authoritative model snapshots. Main still seeds unknown child model and effort from the parent and only changes late model data in memory, so file-defined agents and late client corrections remain wrong. The diff removes those guesses and adds hook-based effort plus task.updated corrections.
Recommendation. Keep open: partial fix. Keep the merged early-model buffer and review the remaining effort and late-correction behavior.
Request. Include line counts for untracked files in status details without changing the real Git index.
Audit finding. Main obtains counts from git diff HEAD and fills missing status paths with zero counts, so the defect remains. The proposal adds a bounded temporary-index probe with failure fallbacks and tests for nested files, required filters, and size limits. It still uses display-format numstat parsing and must be coordinated with the path parser fix.
Recommendation. Keep open: work remains. Combine the untracked-count probe with the NUL-separated path handling proposed in PR 8310.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Force-stop synchronous login-shell probes when their five-second timeout expires.
Audit finding. The shared login-shell probe still uses execFileSync with a timeout but no killSignal, so a shell that ignores SIGTERM can hold startup. Recent PATH-cache and Windows-shell changes do not alter this call. The PR adds SIGKILL only to the POSIX environment probe and leaves successful probes and Windows PowerShell behavior unchanged.
Recommendation. Keep open: work remains. Review the timeout option and verify a shell that traps SIGTERM cannot hold the probe open.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Distinguish hidden, disabled, ambiguous, and missing preview click targets.
Audit finding. Main still collapses hidden and disabled element lookups into notFound. The complete PR carries named failure kinds through desktop errors and the web response without exposing locator text. It does not meet the related issue request to include the selector, and should not close that issue as fixed.
Recommendation. Keep open: work remains. Review the typed click errors while leaving the selector-disclosure request open.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Terminal selection should scroll faster with drag distance and support block, keyboard, all-text, and command-output selection.
Audit finding. Main still autoscrolls one row per 80 milliseconds and clears the selection when replayed terminal state appends output. The PR changes that behavior and adds the four selection paths, with alternate-screen and mouse-tracking guards. This is broader than the recent clipboard fixes and needs one integrated input test that includes live output and an OSC 133 shell.
Recommendation. Keep open: work remains. Verify the complete selection flow in a real terminal and keep a short drag-autoscroll recording.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Expose descriptor-backed reasoning and service-tier choices in the SwiftUI composer.
Audit finding. The experimental parent still lacks FeatureComposerTraitsControl, while this diff adds generic select and boolean sections and preserves hidden prompt-injected values. Its tests cover inherited selections, defaults, and dispatch to a remote environment. This targets the unmerged SwiftUI parent, not the shipped React Native composer.
Recommendation. Keep open: work remains. Review the traits change for the experimental SwiftUI branch after confirming the parent direction.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Observed GitHub metadata. GitHub state OPEN. Author saphid. Updated 2026-08-31T09:53:36Z. Draft no. Target t3code/rebuild-mobile-app-swift. Head 2524de5c55914a9364c56f1dcff7c5953b74ea98. Branch feat/issue110-reasoning-selector. Size +736 / -7, 6 files, 9 commits. Mergeability MERGEABLE. Merge state CLEAN. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Request. Workspace listings should not fail when distinct filesystem paths collapse to the same trimmed wire path.
Audit finding. Main still maps raw file and directory paths without trimming and deduplicating the emitted values. The PR prevents duplicate response paths and gives directories priority, but it intentionally hides colliding filenames instead of making whitespace-bearing paths addressable. That matches its narrow list-failure scope, not full support for those filenames.
Recommendation. Keep open: work remains. Review the deduplication fix while documenting that colliding whitespace filenames remain inaccessible.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Carry Bedrock credentials from the desktop login shell to the server and WSL.
Audit finding. Current shell hydration does not restore the AWS and Claude Bedrock variables added by this diff, and WSL forwarding only includes the two generic API keys. The merged Bedrock authentication recognition does not supply missing GUI-launch credentials. This is a desktop environment fix, not a fix for managed accounts whose initialization omits auth information.
Recommendation. Keep open: work remains. Run the focused desktop shell and WSL environment tests on the current branch.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. GUI launch and Windows-to-WSL execution were not reproduced. Latest-head required checks are not reported: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Use the remote HEAD instead of a stale local origin/HEAD to identify the default branch.
Audit finding. Main still resolves the default through a local symbolic ref and caches that answer. PR 7078 improves worktree fallback selection but does not query the remote HEAD, so renamed default branches remain distinct work. The diff adds a cached five-second ls-remote probe with offline fallback and focused Git tests.
Recommendation. Keep open: work remains. Review the new network probe on the local-first status path and retain the stale-HEAD tests.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Show completion age rather than last-update age on rich SwiftUI Home rows.
Audit finding. The experimental parent still groups done and ready rows under updatedAt and parses invalid completion dates through the fallback date parser. The patch uses a valid latest-turn completion timestamp, preserves the existing fallback, and updates UIKit and SwiftUI accessibility values. It is independent of the React Native unread-Done proposal.
Recommendation. Keep open: work remains. Review completion-age behavior in the experimental SwiftUI parent.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Observed GitHub metadata. GitHub state OPEN. Author saphid. Updated 2026-08-30T02:31:48Z. Draft no. Target t3code/rebuild-mobile-app-swift. Head 8c177b5d97e5869267a2cd959fc395612b5b8254. Branch feat/issue119-done-for-duration. Size +195 / -10, 5 files, 2 commits. Mergeability MERGEABLE. Merge state CLEAN. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Check: GitHub check. Current head 8c177b5d: Test=SUCCESS, Check=SUCCESS, Mobile Native Static Analysis=SKIPPED, Release Smoke=SUCCESS. SwiftUI contract fixtures and native tests passed.
Limits. The body says exact-head UI evidence is not yet attached. Mobile Native Static Analysis was skipped on the current head; no passing native static result is claimed.
Request. Limit preview_snapshot metadata so a large accessibility tree cannot fill the agent context.
Audit finding. Main still serializes all snapshot metadata into both MCP text and structuredContent with no byte budget. The complete PR caps metadata at 64,000 bytes, first removes the accessibility tree, and then emits a smaller summary. The slim-snapshot PR changes defaults, so it complements rather than proves this transport bound is already present.
Recommendation. Keep open: work remains. Keep one metadata-size bound and review it together with the slim-snapshot defaults.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Distinguish remote and WSL projects in the sidebar project filter and its selected trigger.
Audit finding. Main shows only the project name in the sidebar filter trigger and has no RemoteProjectBadge in its project rows. The merged new-thread location change applies to the command palette, not this sidebar control. The diff handles both remote server and desktop-local sandbox labels without changing project identity.
Recommendation. Keep open: work remains. Review the sidebar badge on same-named local, remote, and WSL projects.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Merged pr: PR #7392. Labels the separate new-thread picker only.
Limits. Current-head required checks were not reported: Check, Mobile Native Static Analysis, Release Smoke, Test. GitHub reports merge conflicts with main on the collected head.
Request. Mark experimental SwiftUI Dev and Test builds in the Home title.
Audit finding. The experimental parent still shows the ordinary T3 Code Home label. The diff adds an optional T3BuildChannel plist value and leaves undeclared or unknown builds unchanged, without changing signing or bundle identity. The choice is whether to support this build-channel marker in the experimental client.
Recommendation. Keep open: decision needed. Confirm the build-channel marker policy for the SwiftUI parent.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Observed GitHub metadata. GitHub state OPEN. Author saphid. Updated 2026-08-30T03:49:35Z. Draft no. Target t3code/rebuild-mobile-app-swift. Head 0bb6321643348ddba4efa68bd385c03961544398. Branch feat/issue121-home-channel-title. Size +129 / -1, 5 files, 1 commits. Mergeability MERGEABLE. Merge state CLEAN. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Request. Avoid scanning every installed font for monospace classification when Appearance opens.
Audit finding. FontFamilyPicker still filters the full installed catalog through isMonospaceFamily during render, and the probe still has no result cache. The patch validates only the selected font and its latest cache changes address the review findings. Required checks pass, but the human approval predates the last cache fix and the report does not establish Electron performance with a large catalog.
Recommendation. Keep open: work remains. Review the final cache change and verify selected-font validation in Electron with a large installed-font catalog.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Accept partial-clone annotations when reading repository identity from git remote -v.
Audit finding. Main still requires the fetch marker to end the line, so a trailing blob filter annotation prevents identity detection. Recent identity caching work leaves that parser unchanged. The two-file diff accepts bracketed annotations and adds a real-Git promisor-remote test.
Recommendation. Keep open: work remains. Run the focused identity test on current main and request maintainer review.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Include the failed command name and reason in CommandResolutionError diagnostics.
Audit finding. CommandResolutionError on main still has no message getter. The recent Windows PATH fixes change resolution behavior but leave this tagged error unchanged. The four-line diagnostic change remains independent of those fixes.
Recommendation. Keep open: work remains. Review and merge the message getter after current checks pass.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Limit desktop quit waiting when a pooled backend never completes shutdown.
Audit finding. DesktopApp still stops every pool instance without a timeout before marking shutdown complete. The update-install path already uses a five-second stop budget, but that does not protect normal quit. Closing the window before cleanup can hide this wait without removing it.
Recommendation. Keep open: work remains. Apply the same bounded stop to normal quit and verify a hung backend does not block app exit.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Add independent text and code sizing with ordered settings saves in SwiftUI.
Audit finding. The experimental parent has no textSize or codeSize preferences and saves settings without updating the cached NativeFeatureClient snapshot. The diff applies a window Dynamic Type offset, code-only offsets in hosted content, and serial settings writes with durable rollback. This is not just a settings control change, so cache republish and save failure behavior must stay in its review scope.
Recommendation. Keep open: work remains. Review the sizing and persistence change against the current SwiftUI parent and obtain current-head UI evidence.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Observed GitHub metadata. GitHub state OPEN. Author saphid. Updated 2026-08-28T13:04:56Z. Draft no. Target t3code/rebuild-mobile-app-swift. Head 0eede8d14d22aec0f77f1cf08a90d8bb8aba7346. Branch agent/swiftui-text-code-sizes. Size +780 / -27, 15 files, 1 commits. Mergeability MERGEABLE. Merge state CLEAN. Target is not main. Metadata snapshot 2026-09-01T11:57:41.491617+00:00.
Check: GitHub check. Current head 0eede8d1: Test=SUCCESS, Check=SUCCESS, Mobile Native Static Analysis=SKIPPED, Release Smoke=SUCCESS. SwiftUI contract fixtures and native tests passed.
Limits. Current-head UI evidence is pending according to the body. Mobile Native Static Analysis was skipped on the current head; no passing native static result is claimed.
Request. Keep worktree branch refreshes from fetching or pruning tags under the user's Git configuration.
Audit finding. Main fetchRemote still invokes git fetch --quiet without either tag override. Other fetch helpers use --no-tags, but they are separate paths and do not fix worktree preparation. This diff adds both overrides and an integration test that preserves a moved local tag while updating the tracking branch.
Recommendation. Keep open: work remains. Review the two-file fetch fix and rerun its focused Git integration test on current main.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Windows desktop must forward configured Bitbucket credentials into its managed WSL backend.
Audit finding. Main's WSL forwarding allowlist still contains only the two provider API keys. The PR adds all three supported Bitbucket variables and checks that values reach the child environment while WSLENV contains names only. It does not address macOS shell-profile credential discovery, which is a separate path.
Recommendation. Keep open: work remains. Review the three-variable allowlist addition with the existing WSL environment test.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Add an external-browser action for HTML and PDF file links in chat.
Audit finding. Main still exposes only the integrated-browser helper for file chips. The full PR reserves a web tab before requesting a signed URL, closes it on failure, and uses the desktop shell on Electron. It keeps the selected thread environment as the asset host and does not change the primary file action.
Recommendation. Keep open: work remains. Review the file-chip action and verify its reserved-tab flow against a remote environment.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Show an unread completion before passive monitoring in sidebar status labels.
Audit finding. Main still chooses Monitoring before Done in the default sidebar and before Completed in the legacy pill and project priority table. Server-side settlement does not change that client label order. The separate opacity fix restores contrast but cannot make the completion label visible.
Recommendation. Keep open: work remains. Review the completion-before-monitoring order in default rows, legacy rows, and project rollups.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Integrated terminals should advertise truecolor so terminal applications emit 24-bit colors.
Audit finding. Merged PR #5693 supplies TERM=xterm-256color on Windows, but main still does not set COLORTERM. This PR adds a missing-or-empty COLORTERM default in NodePtyAdapter and preserves caller values. The 256-color change is not a substitute for the truecolor capability signal.
Recommendation. Keep open: work remains. Review the COLORTERM default and verify truecolor output from a TUI.
Request. Fail mobile Kotlin lint when ktlint or detekt is not installed.
Audit finding. The current script warns and exits successfully when either Kotlin tool is missing. The patch changes that branch to a typed failure when Kotlin sources exist and retains the existing SwiftLint skip. Focused tests cover one and two missing Kotlin tools.
Recommendation. Keep open: work remains. Review and run the script tests with missing-tool cases.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Check: PR #7682. Current head e185192a: required checks are absent from canonical latest-commit metadata: Test, Check, Mobile Native Static Analysis, Release Smoke.
Limits. Required CI checks are not present on the current head in canonical check metadata.
Request. Use Claude CLI auth status to detect logged-out first-party accounts.
Audit finding. Current provider status marks every returned capability object authenticated, including one with no account fields. This diff adds a version-gated auth status command and handles logged-out results separately, but it does not run when the capability probe itself fails. The later auth proposals overlap this status logic and add different failure coverage.
Recommendation. Keep open: work remains. Select one auth-status implementation with logged-out, failed-probe, and external-backend tests.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Desktop port selection must exclude ports already listening inside its chosen WSL distro.
Audit finding. Both primary and secondary port scans on main still test only Windows sockets. This PR reads the distro listener tables once before choosing the shared backend port, which avoids the known Windows-versus-WSL port namespace mismatch. It deliberately falls back when probing fails and cannot prevent a port being taken between probe and bind.
Recommendation. Keep open: work remains. Verify primary and secondary WSL startup while a separate distro service already holds the default port.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Explain port bind failures and identify a recorded T3 server on that port.
Audit finding. Main still launches the HTTP server without an EADDRINUSE translation. The patch adds real occupied-port tests and preserves other bind failures, but its recorded PID is not a live ownership check. Desktop port refusal and server-directory locking are separate open changes.
Recommendation. Keep open: work remains. Review the error wording and run the focused bind tests on the current base.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Compare Homebrew provider installs with the version available through Homebrew.
Audit finding. Main can choose brew upgrade, but resolveLatestProviderVersion still always reads npm for packages. Thus a delayed cask update can leave an update notice that brew cannot satisfy. The PR adds separate Homebrew cache keys and cask/formula lookup tests, while intentionally retaining npm for third-party taps.
Recommendation. Keep open: work remains. Review version-source selection with the installer detection work.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Keep unread or newly woken in-flight sidebar rows at full opacity.
Audit finding. Main still exempts unread rows from text recession but applies whole-row opacity to every inactive in-flight row. Server-side settlement and recent sidebar ordering changes did not remove that second dimming path. The separate completion-label patch does not fix contrast.
Recommendation. Keep open: work remains. Review the single attention predicate and retain the separate completion-label work.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Force nonverbose Claude metadata output so the decoder receives one JSON object.
Audit finding. Main only overrides thinking, fast mode, and ultracode settings, leaving the user verbose setting active. The diff adds verbose:false to every metadata call and keeps the decoder strict, so title, branch, commit, and PR generation share the fix. No merged text-generation change in the inspected history pins this setting.
Recommendation. Keep open: work remains. Review and run the focused Claude text-generation tests alongside the isolation changes.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Remove a one-pixel seam where the titlebar fade mask meets the solid mask.
Audit finding. Main still sizes the gradient and solid mask to meet at exactly the same boundary. The later fade-height merge renamed and reduced the height but did not add overlap, so it does not establish that fractional-scale seams are fixed. The proposed one-pixel overlap still addresses a distinct mechanism.
Recommendation. Keep open: work remains. Verify the overlap at fractional display scaling on the current titlebar fade.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Avoid repeating a command as tool detail when the provider adds its tool-name prefix.
Audit finding. Main compares detail directly with the command preview and does not strip a matching tool-name prefix. The current server projection also omits toolName from ordinary command payloads, so the client-only part would not fix the full path. The patch correctly needs both projection and comparison changes.
Recommendation. Keep open: work remains. Review the server projection and client comparison together with a prefixed command payload.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Repair missing execute bits on bundled resource-monitor binaries after npm installation.
Audit finding. Main still rejects a bundled binary with no execute bits and has no repair path. This patch limits chmod to bundled candidates and leaves override paths unchanged. Open PR 8650 addresses future package metadata instead of repairing installed files, so it is related work rather than an exact duplicate.
Recommendation. Keep open: work remains. Coordinate the installed-file repair with the separate package-mode correction.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. No installed npm package was tested during this read-only audit. Required CI checks are absent at the reviewed head: Test, Check, Mobile Native Static Analysis, Release Smoke.
Request. Make unchecked switches visible when an imported theme has a transparent input color.
Audit finding. Both Switch and MenuCheckboxItem on main still use data-unchecked:bg-input. The two-line patch changes both tracks to muted-foreground at 30% opacity, without changing checked states, so the reported custom-theme defect remains uncovered.
Recommendation. Keep open: work remains. Review the two token changes with custom-theme contrast evidence.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Request. Add an explicit command to remove unused pinned service runtimes.
Audit finding. Main installs immutable runtimes and exposes install, update, uninstall, and status, but no prune command. The diff protects the active and rollback versions, ignores incomplete installs and symlinks, and adds dry-run and deletion tests. This remains a server maintenance feature for systemd and launchd services, not a landed cleanup fix.
Recommendation. Keep open: work remains. Review the candidate-selection and concurrent-update boundary before merging.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.
Limits. Required current-head checks are absent: Test, Check, Mobile Native Static Analysis, Release Smoke. GitHub reports merge conflicts with the current base.
Audit finding. Main drops a skill as soon as the YAML parser throws, even when the CLI reports the same skill as a command. The diff adds a limited scalar recovery path and tests comments plus broken structural fields. The merged workspace discovery change does not alter this parser, so this compatibility gap remains.
Recommendation. Keep open: work remains. Review the lenient fixtures against Claude CLI parsing and run the skill-discovery tests.
Confidence high. Release: Not applicable. PR readiness: Needs maintainer review.